CAEPR Working Paper

#002-2010

Estimating the Effects of Dormitory Living on Student

Performance
Pedro de Araujo
Colorado College

James Murray
University of Wisconsin - La Crosse
February 9, 2010

This paper can be downloaded without charge from the Social Science Research Network
electronic library at: http://ssrn.com/abstract=1555892.

The Center for Applied Economics and Policy Research resides in the Department of Economics
at Indiana University Bloomington. CAEPR can be found on the Internet at:
http://www.indiana.edu/~caepr. CAEPR can be reached via email at caepr@indiana.edu or
via phone at 812-855-4050.

©2008 by NAME. All rights reserved. Short sections of text, not to exceed two paragraphs, may
be quoted without explicit permission provided that full credit, including © notice, is given to
the source.

Electronic copy available at: http://ssrn.com/abstract=1555892

Estimating the Eects of Dormitory Living on Student
Performance∗
† ‡
Pedro de Araujo James Murray

Colorado College University of Wisconsin - La Crosse

February 9, 2010

Abstract
Many large universities require freshman to live in dormitories on the basis that
living on campus leads to better classroom performance and lower drop out incidence.
Large universities also provide a number of academic services in dormitories such as
tutoring and student organizations that encourage an environment condusive to learn-
ing. A survey was administered to college students at a large state school to determine
what impact dormitory living has on student performance. We use a handful of instru-
mental variable strategies to account for the possibly endogenous decision to live on
campus. We nd a robust result across model specications and estimation techniques
that on average, living on campus increases GPA by between 0.19 to 0.97. That is,
the estimate for the degree of improvement to student performance caused by living on
campus ranges between one-fth to one full letter grade.

Keywords : Student performance, dormitory, cross-section analysis, regression, instru-

mental variables.
JEL classication : C13, C21, I21.

∗
We thank the participants of the 2009 Sourthern Economics Conference for their useful comments. All
errors are our own.
†
Department of Economics and Business, 14 E. Cache La Poudre Street, Colorado Springs, CO 80903.
Phone : (719)389-6470. E-mail : Pedro.deAraujo@ColoradoCollege.edu.
‡
Department of Economics, 1725 State Street, La Crosse, WI 54601. Phone : (608) 406-4068
E-mail : murray.jame@uwlax.edu

Electronic copy available at: http://ssrn.com/abstract=1555892

 Eects of Dormitory Living on Student Performance 1

1 Introduction
It is widely believed that students can acquire academic benets from living on campus.
So much so, that many colleges and universities, ranging from small liberal arts colleges to
large state universities, require students to live on campus during their rst year with few
exceptions. Typically, students exempt from such a policy include students over the age of
25, students that are married and/or have children, and students in the military. It has been
suggested that living on campus causes students to be less likely to drop out or transfer, more
likely to make academic progress, and more capable of achieving a high level of academic
performance.
Despite these common perceptions, there are inherent diculties in estimating the impact
living on campus has on student performance. For the schools that require freshman students
to live on campus, there is no eective control group. Typically, freshman students that
do not live on campus share other features with each other that set them apart as non-
traditional students. Many of these schools only require freshman to live on campus, so one
might be tempted to compare the academic performance of sophomores that live on campus
versus o campus. The problem with such a strategy is that assignment of a sophomore
student to on-campus versus o-campus housing is not random. Rather it is the choice of an
individual student, and this choice likely depends on his or her perceptions of the academic
benets he or she received while living on campus during the freshman year. Students that
did receive academic benets are more likely to stay on campus, while students that did not
are more likely to move o campus. This is known as a self-selection problem in regression
analysis. As a consequence the explanatory variable, living on campus, is jointly determined
with the dependent variable, some measure of academic performance. This leads to biased
and inconsistent estimates of the impact living on campus has on academic performance.
Comparing the performance of on-campus versus o-campus students at schools that do
not have an on-campus housing requirement can also lead to a self-selection problem. The
students (or parents of students) that perceive obtaining the greatest benets of living on
campus are more likely to choose to do so. If the selection problem is ignored, it is not clear
in which direction this would bias the results. It is possible the students who perceive the
most benet from living on campus are most represented by motivated students that plan on
reaping these benets. Or in the case the parents made this decision, the students living on
campus may be more represented by parents that have high educational attainment and/or
high expectations for their children. In these cases, ignoring the self-selection problem would
lead to an over-estimate of the true impact living on campus has on student performance.
On the other hand, it is possible that the students (and/or parents of students) that perceive
having the most to gain are largely represented by the students that are in most need of the
possible benets for successful academic performance. In this case, ignoring the self-selection
problem would lead to an under-estimate of the true eect.
The purpose of this paper is to quantify the impact living on campus has on student per-
formance, as measured by grade point average (GPA), taking into careful consideration the
possible self-selection problem. We use two instrumental variables, (1) distance of hometown
from campus and (2) a dummy variable for being denied housing due to space limitations.
These variables inuence whether or not a student lives on campus, but are otherwise un-
correlated with outcomes for academic achievement. This study uses survey data collected
 Eects of Dormitory Living on Student Performance 2

from sophomore students and above at Indiana University Purdue University Indianapolis
(IUPUI) in Fall 2008 on current and past living situations, past semester and cumulative
1
GPA, and student background characteristics.
Our main nding is that living on campus (whether only in past semesters or currently)
increases a single semester and cumulative GPA by about a half of a letter grade. These
results are robust to dierent instrumental variable strategies and dierent estimation meth-
ods. Since self-selection problems are taken into account, these results have immediate policy
implications: colleges and universities can expect that encouraging students to live on cam-
pus, even if just for a few semesters, on average will result in better academic performance
throughout the students' college careers.

2 Related Literature
A substantial body of work explores the determinants of academic achievement with a sub-
set of this literature focused on the impact students' residences. The ndings are somewhat
mixed. Thompson, Samiratedu, and Rafter (1993) nd that freshman students who live on
campus have higher retention, a greater degree of academic progress, and higher academic
performance. Delucchi (1993) examines a college town where most students who live o
campus are still in close walking distance of their classes and university resources and nds
no statistically signicant dierence in academic achievement between students that live on
campus and o campus. Flowers (2004) focuses exclusively on African American students
and nds that living in dormitories positively inuenced measures of personal and social de-
velopment skills that he suggests are essential for successful academic achievement. Pike and
Kuh (2005) nd residence is important when they examine the experiences of rst-generation
students. They nd that these students' characteristic low levels of academic engagement is
a function living o-campus in addition to having lower educational aspirations.
Pascarella, Bohr, Nora, Zusman, Inman, and Desler (1993) nd that students who live
on campus achieve larger gains from college when in comes to measures of critical think-
ing and cognitive skills, but nd less impact when it comes to direct measures of reading
comprehension and mathematical skills. This may suggest that the measurable gains to
academic performance are not immediate, but may be delayed for a number of semesters.
If living on campus causes growth in a student's critical and cognitive thinking skills, one
would expect these are permanent improvements which should pay o in higher academic
achievement throughout the student's career. For this reason, in the methodology section
below, we include in the analysis how living on campus at any time in the past may aect
cumulative GPA.
Further research on this subject is essential for two reasons. First, most of these studies
are somewhat dated; some results may be primarily applicable to only to situations that ex-
isted when the studies were conducted, some of these ten to twenty years ago. Secondly, these
studies do not explicitly address the self-selection problem discussed above, and therefore
may not have relevant policy implications.
The rst of these criticisms, that simply too much time has progressed since many of

1 Freshman
students were excluded because the survey was administered in the Fall semester, a time in
which freshman students are in their rst semester and do not yet have any academic records.
 Eects of Dormitory Living on Student Performance 3

these studies were completed, may seem somewhat weak at rst, but it is a particularly
important concern when it comes to education. As Pascarella and Terenzini (1991) point
out, student characteristics and features of higher education have changed signicantly over
the years. The diversity of ethnic, cultural, and socio-economic backgrounds of students have
expanded over the years; educational technology has developed substantially; and teaching
and learning techniques have evolved. In fact all these characteristics of students and college
learning have evolved tremendously since Pascarella and Terenzini pointed them out in 1991.
We should expect the dynamics between academic achievement and students' backgrounds,
students' peers, and institutional characteristics have also changed.
Addressing the second criticism above is the main contribution of this paper. Using
instrumental variables we account for the possible self-selection problem, so statistically
signicant results from this study do indicate that living on campus causes improved student
performance. Accounting for the self-selection problem, we can rule out that these eects can
be explained by more academically capable and/or more highly motivated students being
more likely to choose to live on campus than other students.
Besides literature focusing on specically on students' residences, there is other research
on academic performance that shed some light as to why students who live on campus may
perform better. Students that live on campus may be more likely to benet from university
provided resources. Toutkoushian and Smart (2001) nd empirical evidence that increases
in institutional spending leads to improvements in students' learning. However, they also
nd that increases in funding to academic support does not necessarily improve learning,
suggesting that even non-academic resources on campus create an environment that fosters
learning and good study habits. Such resources may include services provided in dormitories,
but they also likely include resources that on-campus students may be more likely to benet
from, such as computer and information technology, university clubs, university sponsored
varsity and intramural sports, exercise facilities, and other extra-curricular activities.

3 Data and Methodology

Data was collected in Fall 2008 from students pursuing four-year baccalaureate degrees from
IUPUI. This population consists of 30,000 students, approximately 19,700 students under the
age of 25 (the population that would most likely consider on-campus housing), and an on-
campus housing capacity of only 1,107. Since there is such a small availability of on-campus
housing, living on campus is not required. Most students do not live on campus, and those
that do are primarily freshman. An electronic survey that takes about 15 minutes to complete
was sent to 6,000 representative undergraduate students that were sophomore level or above
that asked them a variety of questions on background characteristics, living situations, social
habits, study habits, university involvement and academic performance. Although it is not
a perfect measure of academic ability or the personal and educational gains from college,
academic performance is measured by students' last semester and cumulative grade point
average (GPA). Of the students surveyed, 363 completed the questionnaire. Approximately
15% of the students who completed the survey had lived on campus during some part of
their time at IUPUI which is consistent with the population of IUPUI students.
 Eects of Dormitory Living on Student Performance 4

The structural equation for the relationship we investigate is given by,

GP Ai = α + βDORMi + Xi0 Ω + i , (1)

where subscript i denotes an individual student; GP Ai is an individual student's cumulative

GPA or previous single-semester GPA, each examined in turn; DORMi is a dummy variable
that is equal to 1 if student i has ever lived on campus, or if a student lived on campus last
semester, each examined in turn. The vector Xi includes controls including gender, parents
income (P IN C ), SAT/ACT percentiles (T EST ), the total number of semesters the student
has been enrolled for classes at IUPUI (T SEM ), a dummy if the student is older than 25
(N T S ), and an interaction between T SEM T EST .2 The total number of credit hours a
and
student was taking in the previous semester (T N C ) is also included in the regressions where
the dependent variable is previous semester GPA.
When measuring parents' combined income using a survey administered to students, the
students were asked to identify one of several income range categories that described their
parents income. Each category included a range of $20,000, and the nal category was More
than $200,000. The PINC variable is coded using the midpoint from each category. In the
event the student selected the nal category, the midpoint from the second highest category
was entered for PINC, and a dummy variable, P IN C _d was set equal to 1. The interaction
term between ACT/SAT test and total semesters in school is included to account for the
possibility that performance on standardized tests taken in high school are weaker predictors
of success in college the longer the student has been in school.
We estimate three specications of the model. In Specication 1 we regress cumulative
GPA on ever having lived on-campus. This specication is motivated by the literature cited
above that suggests many of the benets one obtains from living on campus should be long
lasting. These benets include growth in cognitive thinking, growth in critical thinking, and
improved interpersonal skills. If students acquire these skills while living on campus during
their freshman year, but subsequently move o campus - which is typical at IUPUI - we
should still see benets to these students cumulative GPA during their junior and senior
years. Specication 2 is similar except we regress only the previous semester GPA (not
cumulative - only the average for the courses taken in the previous semester) on ever having
lived in a dorm. Finally, Specication 3 is most like the previous literature. Here we regress
semester GPA on whether or not a student is currently living in the dorm. This specication
captures instantaneous benets to living in a dorm. Such benets may be greater access to
academic resources such as computer technology, libraries, tutors, and even professor oce
hours.
Diculty arises if the self-selection problem causes E(i |DORMi ) 6= 0, that is if the
choice of living on-campus is endogenously determined with students' academic success.
This happens when characteristics that inuence academic success, like motivation, parental
inuence, and incoming ability, also inuence the decision to live on-campus. One way to deal

2 Only a dummy for being over the age of 25 and not actual age was included in the regression for two
reasons. First, any linear relationship between age and academic performance likely diminishes by the time
students reach age 25, an age which typically categorizes a person as a non-traditional student. Secondly,
for students under the age of 25, the number of total semesters the student had attended IUPUI and their
age was very highly correlated.
 Eects of Dormitory Living on Student Performance 5

with an endogenous regressor is to use the control function approach (Cameron and Trivedi
[2005]). That is, as long as a subset of our control variables that include students' abilities
and motivation capture the endogenous decision to live on campus, the OLS estimates are
consistent. There are two such variables in the model, ACT/SAT percentiles and parents
income, so it is possible that OLS is appropriate. We will formally test this possibility in
the next section.
In the case we still have an endogeneity problem, we employ two instrumental variables:
distance of hometown from campus (DIST ) and a dummy variable for whether or not
students were denied access to dormitories due to lack of space (DEN ). When dealing
with more instrumental variables than endogenous variables, there are multiple estimation
strategies to consider. In addition to estimating the three specications of the model using
OLS, we employ the three instrumental variable estimation techniques. The rst strategy
(IV) we use only distance from campus as an instrumental variable and conduct a standard
just-identied instrumental variable regression. The second strategy we consider is to use
both instruments and estimate the model using generalized method of moments (GMM), a
strategy which identies the coecients using the moment condition that both instrumental
variables are orthogonal to the error term. The benet of this technique is that while it uses
both instruments and assumes the structure in equation (1), it does not impose any other
structure on the error term, such as normality or homoskedasticity. The drawback inherent
with GMM approaches in general, is that without further structure it can be dicult to
obtain statistical signicance. Finally, we estimate the model using a two stage maximum
likelihood (MLE) approach. In the rst stage we regress the endogenous variable, DORM ,
against the instruments and control variables, which uses a probit since it involves a dummy
dependent variable. The second stage regresses GPA on the predicted values of DORM from
the rst stage, along with the control variables. This nal approach imposes restrictions of
normality and homoskedasticity on the error term.

4 Results
3
The results for Specications 1, 2, and 3, are given in Tables 1, 2, 3, respectively. We
have a quite robust result: with the exception of the just-identied IV technique, all other
estimation strategies result is a positive, statistically signicant eect of living on campus
on student performance, for every specication in the model. The results in Table 1 suggest
the increase of cummulative GPA obtained from ever having lived on campus ranges from
0.2 to about 0.5. This is a large enough increase to actually recieve a dierent letter on
one's transcript (for example a B versus a B-, or a B+ versus a B-). Other controls that are
statistically signicant include GEN DER, women on average perform better than men; and
T EST , successful performance on standardized tests in high school helps explain success in
college. Though, while the coecient on T EST is statistically signicant, it has a relatively

3 Weakinstrument tests (Stock and Yogo test and joint F test) have been performed and there is enough
evidence that both instruments are jointly not weak in specication 1. This is not the case in specications
2 and 3 (there is evidence that they are not weak only if we allow for a 20 to 30% relative OLS bias in the
coecient estimate). We show later these are the specications where there is the least amount of evidence
of endogeneity.
 Eects of Dormitory Living on Student Performance 6

small impact on GPA.

Comparing results across estimation techniques illustrates the consequences of ignoring
the self-selection problem. When accounting for endogeneity using GMM and MLE, the
predicted increase on GPA is close to 0.5. When ignoring the problem, the point estimate
decreases to around 0.2. This suggests the students that choose to live on campus are more
highly represented by students with lower incoming academic ability, i.e. students that can
get the most benet from living on campus. Ignoring the endogenous decision to live on
campus results in these students bringing down the average performance of those living in
the dorm, leading to a possibly incorrect conclusion that living in dorms has a relatively
small impact on GPA.
The results for Specication 2 in Table 2 largely mirror the results for Specication
1. Having ever lived on campus causes an increase in both semester GPA and cumulative
GPA. Table 3 shows the results for Specication 3, the instantaneous eects of living on
campus. The instantaneous eects are quite large. The instrumental variable methods
GMM and MLE reveal that students that are currently living on campus are able to achieve
a GPA that is between 0.7 to 1.0 higher due to living on campus. This may be due to
increased utilization of academic resources provided on campus, positive peer inuences from
other students living in dorms, or the organization, structure, and activities that dormitories
provide their residents to encourage an environment conducive to learning.
In an eort to determine which estimation strategy is most appropriate we conduct
Durbin-Wu-Hausman tests for endogeneity. The null hypothesis for this test is no endogene-
ity, so failure to reject the null hypothesis may imply that a self-selection problem is not
present, or it is accounted for by one or more of the control variables. Failure to reject the
null hypothesis is not very conclusive, however, as it may be simply due to lack of statisti-
cal evidence. Rejection of the null hypothesis implies that endogeneity is a problem, OLS
results are inconsistent, and the instrumental variable strategies are more appropriate. To
determine how well the controls can account for the choice to live on campus, the test is run
under four cases: (1) with all controls included in the model, (2) with all controls except
ACT/SAT test percentiles (T EST ), (3) with all controls except parents income (P IN C and
P IN C _d), and (4) with all controls excluding both ACT/SAT test percentiles and income.
The p-values for the Durbin-Wu-Hausman tests are reported in Table 4. The rst column
shows when using all controls, for all but one specication and estimation strategy choice,
the exogeneity for DORM cannot be rejected. In Specication 1, when excluding T EST
(whether only T EST [column 2], or both T EST and P IN C [column 4]) exogeneity is
strongly rejected. These results suggest a self-selection issue is certainly present, but using
incoming ability as measured by ACT/SAT test scores as a control in the regression may
account for much of the problem. The p-values for Specication 2 and 3 that use single
semester GPA are often not statistically signicant, indicating endogeneity may not be a
problem in these specications.

5 Conclusion
Despite a widely held belief that living on campus helps students perform better in school, a
look at the literature reveals mixed results, and a failure in these studies to account for the
 Eects of Dormitory Living on Student Performance 7

possible endogenous decision students make on whether or not to live on campus. Factors
that inuence how well a student can perform likely also inuence a particular student's
choice on where to live. In addition to estimating a standard model using OLS, we care-
fully account for endogeneity using two instrumental variables: distance of hometown from
campus and a dummy variable for being denied on-campus housing due to space. To ensure
robustness of the results, we estimate the model using three instrumental variable procedures,
and examine the evidence for endogeneity in each. For nearly every estimation strategy and
specication examined, we nd that living on campus does cause an increase in student
performance. We nd evidence for an instantaneous eect. Living on campus during the
semester before the survey was administered caused an average increase in semester GPA
of nearly one full letter grade during that same time. There is also evidence that living on
campus as long term benets, even after students subsequently move o campus. Estimation
results indicate having lived on campus during any time in the past caused an increase in
semester GPA and cumulative GPA of almost half a letter grade. Because care was taken
to account for the self-selection problem, and because the results are robust across specica-
tions and estimation strategies, these ndings have immediate policy implications: requiring
or otherwise encouraging students to live an campus can result in improved academic per-
formance.
 Eects of Dormitory Living on Student Performance 8

References
Astin, A. W. (1993): What Matters in College: Four critical Years Revisited.
Jossey-Boss, San Francisco.
Betts, J. R., and D. Morell (1999): The Determinants of Undergraduate Grade Point
Average: The Relative Importance of Family Background, High School Resources, and
Peer Group Eects, Journal of Human Resources, 34, 268293.
Cameron, A. C., and P. K. Trivedi (2005): Microeconometrics: Methods and
Applications. Cambridge University Press, New York.
(2009): Microeconometrics Using Stata. Stata Press, 1st edn.
Coleman, J. S. (1966): Equality of Educational Opportunity. U.S. Department of Health,
Education, and Welfare, Oce of Education, U.S. Government Printing Oce.
Dellas, H., and P. Sakellaris (2003): On the Cyclicality of Schooling: Theory and
Evidence, Oxford Economics Papers, 55, 148172.
Delucchi, M. (1993): Academic Performance in a College Town, Education, 116.
Epple, D., and R. E. Romano (1998): Competition Between Private and Public
Schools, Vouchers, and Peer-Group Eects, American Economic Review, 88, 3362.
Flowers, L. A. (2004): Eects of Living on Campus on African American Students'
Educational Gains in College, NASPA Journal, 41, 277293.
Hanushek, E. A., J. F. Kain, J. M. Markman, and S. G. Rivkin (2003): Does Peer
Ability Aect Student Achievement?, Journal of Applied Econometrics, 18, 527544.
Heckman, J. J. (1979): Sample Selection Bias as a Specication Error, Econometrica,
47, 153161.
Henderson, V., P. Mieszkowski, and Y. Sauvageau (1978): Peer Group Eects and
Educational Production Functions, Journal of Public Economics, 10, 97196.
Kanoy, K. W., and J. W. Bruhn (1996): Eects of a First-Year Living and Learning
Residence Hall on Retention and Academic Performance, Journal of the Freshman Year
Experience and Students in Transition, 8, 723.
Karemera, D., L. J. Reuben, and M. R. Sillah (2003): The Eects of Academic
Environment and Background Characteristics on Student Satisfaction and Performance:
The Case of South Carolina State University School of Business, College Student
Journal, 37, 298308.
Kowack, K. W., and A. L. Hanson (1985): Academic Achievement of Freshman as a
Function of Residence Hall Housing, NASPA Journal, 22, 2228.
Kuh, G. D., and S. Hu (2001a): The Eects of Student-Faculty Interacation in the
1990s, Review of Higher Education, 24, 309332.
(2001b): The Relationship Between Computer and Information Technology Use,
Selected Learning and Personal Development Outcomes, and Other College
Experiences, Journal of College Student Development, 42, 217232.
Kuh, G. D., C. R. Pace, and N. Vesper (1997): The Development of Process
Indicators to Estimate Student Gains Associated with Good Practices in Undergraduate
Education, Research in Higher Education, 38, 435454.
Lenning, O. T., L. A. Munday, O. B. Johnson, A. R. VanderWell, and E. J.
Brue (1975): The Many Faces of College Success and Their Nonintellective Correlates.
American College Testing Program, Iowa City.
 Eects of Dormitory Living on Student Performance 9

Pascarella, E., L. Bohr, A. Nora, B. Zusman, P. Inman, and M. Desler (1993):

Congnitive Impacts of Living on Campus versus Commuting to College, Journal of
College Student Development, 34, 216220.
Pascarella, E., and P. T. Terenzini (1991): How College Aects Students: Findings
and Insights From Twenty Years of Research. Jossey-Bass, San Francisco.
Pike, G. R., and G. D. Kuh (2005): First- and Second- Generation College Students: A
Comparison of Their Engagement and Intellectual Development, Journal of Higher
Education, 76, 276301.
Sacerdote, B. (2001): Peer Eects with Random Assignment: Results for Dartmouth
Roommates, Quarterly Journal of Economics, 116, 681704.
Sax, L. J., S. K. Gilmartin, and A. N. Bryant (2003): Assessing Response Rates
and Nonresponse Bias in Web and Paper Surveys, Research in Higher Education, 44,
409432.
Schrager, R. H. (1986): The Impact of Living Group Social Climate on Student
Academic Performance, Research in Higher Education, 25, 265276.
Schroeder, C. C., and P. Maple (1994): Residence Halls and the College Experience:
Past and Present, in Realizing the Educational Potential of Residence Halls, ed. by
C. C. Schroeder, and P. Maple, pp. 107132. Jossey-Bass, San Francisco.
Thompson, J., V. Samiratedu, and J. Rafter (1993): The Eects of On-Campus
Residence on First-Time College Students, NASPA Journal, 31, 4147.
Toutkoushian, R. K., and J. C. Smart (2001): Do Institutional Characteristics
Aect Student Gains from College?, Review of Higher Education, 25, 3961.
Wooldridge, J. M. (2002): Econometric Analysis of Cross Section and Panel Data. MIT
Press, Cambridge, MA.
Zimmerman, D. J. (2003): Peer Eects in Academic Outcomes: Evidence From a
Natural Experiment, Review of Economics and Statistics, 85, 923.
 Eects of Dormitory Living on Student Performance 10

Table 1: Specication 1 - Cumulative GPA / Having Ever Lived On-Campus

OLS IV GMM MLE
DORM_E 0.210** 0.312* 0.448*** 0.431***
[0.087] [0.187] [0.140] [0.156]
GENDER -0.200** -0.214** -0.234*** -0.220***
[0.085] [0.088] [0.088] [0.085]
PINC -0.001 -0.001 -0.001 -0.001
[0.0009] [0.0009] [0.0009] [0.0009]
PINC_d 0.065 0.042 -0.002 0.023
[0.183] [0.200] [0.198] [0.183]
NTS 0.027 0.056 0.078 0.084
[0.137] [0.152] [0.150] [0.140]
TEST 0.004** 0.003* 0.003* 0.003*
[0.001] [0.001] [0.001] [0.001]
TSEM -0.010 -0.011 -0.011 -0.013
[0.014] [0.016] [0.016] [0.014]
TEST_TSEM -0.0001 -0.0008 -0.00006 -0.00005
[0.0002] [0.0002] [0.0002] [0.0002]
Instruments - DIST DIST DIST
- - DEN DEN
N 227 226 226 226
Wald Chi n.a. 29.42*** 41.81*** 32.23***
F-stat 3.78*** - - -
2
R 0.122 0.113 0.088 -

Note: * signicant at 10%; ** signicant at 5%; *** signicant at

1%. Standard errors in brackets.
 Eects of Dormitory Living on Student Performance 11

Table 2: Specication 2 - Semester GPA / Having Ever Lived On-Campus

1
OLS IV GMM MLE
DORM_E 0.185* 0.221 0.416** 0.410**
[0.095] [0.289] [0.212] [0.166]
GENDER -0.247** -0.254*** -0.262 -0.257***
[0.100] [0.099] [0.100] [0.091]
PINC -0.001 -0.0009 -0.0007 -0.0009
[0.001] [0.001] [0.001] [0.001]
PINC_d -0.006 -0.015 -0.037 -0.038
[0.216] [0.208] [0.205] [0.194]
NTS 0.216 0.225 0.253* 0.259*
[0.013] [0.141] [0.137] [0.146]
TEST 0.002 0.002 0.001 0.001
[0.002] [0.002] [0.002] [0.002]
TSEM -0.017 -0.017 -0.017 -0.018
[0.022] [0.022] [0.021] [0.016]
TEST_TSEM 0.0001 0.0001 0.0001 0.0001
[0.0002] [0.0002] [0.0002] [0.0002]
TNC 0.007 0.006 0.002 0.003
[0.008] [0.010] [0.009] [0.008]
Instruments - DIST DIST DIST
- - DEN DEN
N 217 216 216 216
Wald Chi - 18.57*** 24.5*** 24.7***
F-stat 2.59*** - - -
2
R 0.096 0.095 0.068 -

Note: * signicant at 10%; ** signicant at 5%; *** signicant at

1%. Standard errors in brackets.
1
OLS is computed using heteroskedastic robust standard errors.
 Eects of Dormitory Living on Student Performance 12

Table 3: Specication 3 - Semester GPA / Semester On-Campus

1
OLS IV GMM MLE
DORM_S 0.303*** 0.490 0.973* 0.693***
[0.096] [0.642] [0.526] [0.201]
GENDER -0.261*** -0.283*** -0.315*** -0.297***
[0.100] [0.104] [0.107] [0.092]
PINC -0.0009 -0.0007 -0.0003 -0.0006
[0.001] [0.001] [0.001] [0.001]
PINC_d -0.010 -0.036 -0.097 -0.055
[0.215] [0.213] [0.222] [0.194]
NTS 0.199 0.215 0.239* 0.227
[0.134] [0.135] [0.134] [0.144]
TEST 0.002 0.001 0.0002 0.0008
[0.002] [0.002] [0.002] [0.002]
TSEM -0.017 -0.018 -0.019 -0.019
[0.022] [0.022] [0.021] [0.016]
TEST_TSEM 0.0001 0.0001 0.0002 0.0002
[0.0002] [0.0003] [0.0003] [0.0002]
TNC 0.008 0.007 0.002 0.005
[0.007] [0.009] [0.009] [0.008]
Instruments - DIST DIST DIST
- - DEN DEN
N 217 216 216 216
Wald Chi - 18.01*** 19.82*** 30.32***
F-stat 2.96*** - - -
R2 0.107 0.099 - -

Note: * signicant at 10%; ** signicant at 5%; *** signicant at

1%. Standard errors in brackets.
1
OLS is computed using heteroskedastic robust standard errors.
 Eects of Dormitory Living on Student Performance 13

Table 4: P-Values from Endogeneity Tests (Null: no endogeneity)

Specication 1 - Cumulative GPA / Having Ever Lived On-Campus
Estimation Strategy Full Model w/o TEST w/o PINC w/o TEST and PINC
IV 0.606 0.074* 0.585 0.042**
GMM 0.366 0.030** 0.368 0.018**
MLE 0.115 0.043** 0.118 0.053*

Specication 2 - Semester GPA / Having Ever Lived On-Campus

Estimation Strategy Full Model w/o TEST w/o PINC w/o TEST and PINC
IV 0.893 0.374 0.801 0.296
GMM 0.670 0.251 0.587 0.201
MLE 0.133 0.182 0.090* 0.139

Specication 3 - Semester GPA / Semester On-Campus

Estimation Strategy Full Model w/o TEST w/o PINC w/o TEST and PINC
IV 0.801 0.288 0.722 0.213
GMM 0.585 0.160 0.586 0.143
MLE 0.072* 0.293 0.071* 0.169

Note: * signicant at 10%; ** signicant at 5%; *** signicant at 1%.